Valuing Free AgentsBy Kevin Pelton
Aug. 8, 2002
One of the trickiest things to do in the modern NBA is value players, particularly free agents. There are a number of different ways that fans mentally go about valuing players. The most frequently used method is to compare the player to others of similar abilities, and see what these other players are making (assuming, of course, they are not on rookie contracts, which are no indicator of a playerís real value). There is a significant problem with this, however; this may shock you, but some players in the NBA are overpaid! There are also those who are underpaid. Say weíre talking about a free agent big man who averaged about 14 points and 6 and a half boards last year. That production could be compared to that of former Seattle SuperSonic and current Boston Celtic Vin Baker . . . who will pull in $12.375 million for his efforts next season. Sounds like a favorable comparison for an agent to make, right?
The answer to players like Baker and Steve Nash (heís the best example I can find off-hand of an underpaid veteran, pulling in less than $6 million for All-Star caliber point play) is to average the similar players. But with limited sample sizes, this might not always work. After all, how many players can you name that are quote unquote ďsimilarĒ in value to any given player? Beyond a handful, youíre stretching.
With that in mind, Iíd like to humbly offer an alternative.
By way of background, last winter in my statistics course at the University of Washington, I was asked to do a simple regression project requiring me to predict a dependent variable based on an independent variable. Naturally, I turned to sports.
My first thought was to use baseball salaries and some statistical performance measure, because baseball has a Ďfreerí market economy than basketball (no salary cap, no maximum salaries, and no Ďexceptionsí). However, this proved difficult for two reasons. First, statistical measures that can be used to compare hitters and pitchers are difficult to come by. The best of these is likely Bill Jamesí Win Shares, but his book listing these had not yet been published when I undertook this project. Second, while salary data for NBA players is fairly easy to find, I had a difficult time tracking down data for lower-level baseball free agents. Without the players at the bottom of the market, it would have been tough to get a complete picture of baseball finances.
After that initial failure, I turned to basketball. Though less ideal in economic terms, the NBA fared much better in amount of effort necessary to complete the project, a key concern, especially given the fact I was quite confident in my ability to get an A with little work. For salary info, I used Patricia Benderís salary listing. For the statistical measure of performance in the past season, I was able to use my own measure, Value Over Replacement Player (VORP). Finding the salary for the 2001-02 season and the VORP for the 2000-01 season for each player who was in the NBA in both seasons and changed teams as a free agent during the summer of 2001 (78 players in total), I quickly finished my project.
After I was finished, however, my interest was piqued. For my short-lived column at ProSportsWriters.net, I was looking at alternative player rating systems to the traditional one, adding good stuff and subtracting bad stuff, most prominently exemplified by the TENdex system created by Dave Heeren. My thought was to rate players based on how organizations themselves rated them, in terms of how much they made as free agents. Instead of VORP, then, I did regression analysis comparing a number of other statistics to the salaries of the free agents I had previously used.
The inherent danger in using the results as a rating system was that salary naturally included components beyond performance in the previous season, notably age, position, and performance two years past and earlier. But I believed that these other factors would more or less average out, making the weights I obtained accurate enough to at least be interesting.
By this summer, the natural application for this formula hit me -- plugging in the same statistics for this summerís free agents to see what they should be worth. A true statistician would immediately have a problem with this application, as it is extrapolating beyond the scope of the input data. In truth, all I should really be able to do is determine what a given player might have made had they been a free agent last summer. But in a frivolous case like this, I donít think predicting 2002 salaries from the 2001 market is a real problem. Normally, inflation would be a significant concern, but with the salary cap actually decreasing this summer and the exceptions remaining fairly close to their value from last summer, I think the values Iíve obtained are reasonable.
Now that I am actually trying to predict salary as opposed to rate players, the other components mentioned above become important. All other things equal, a 25-year-old free agent will make more than a 35-year-old, and a 7-1 center will make more than a 6-2 point guard. To account for this, I added three Ďattributesí to the calculations -- age, experience (more accurately, years since leaving college; I included this mostly to see if there was any difference between players who came out early and those who did not, and account for it if so), and height.
The statistics I used were games played, minutes, field goals attempted, free throws attempted, offensive and defensive rebounds, assists, steals, blocks, turnovers, and points. All of these were raw totals, with my hope that the inclusion of games and minutes would essentially work to recreate the effect of per-game and per-minute averages. In other words, while piling up a lot of points will be a positive, minutes will be a negative, meaning that a player who has scored 1000 points in 2000 minutes will get relatively more credit from his scoring than one who has scored 1000 points in 3000 minutes. The same thinking applies with including field goals attempted and free throws attempted. The one thing I am really unsure about leaving out is some measure of three-point ability; in the end, I decided that it probably wasnít important how points were scored, just that they were scored.
Before I get to the actual ratings, I am personally obligated to make a disclaimer statement. These are very rough estimates at best, and by no means do I necessarily believe what the formula says a given player is worth. Statistics can only explain so much, and they fail to properly account for the man defensive ability of players like Bruce Bowen. There is value, here, however, in my opinion. When the formula says Allan Houston is worth just over $6 million and he gets more than twice that, something is wrong with the Knicksí decision-making. Remember, in the end, that these values are just for recreational use.
2002 Free Agent Rankings
1. $10.90 million - Bonzi Wells, sg, POR
2. $8.58 million - Jeff McInnis, pg, LAC
3. $7.87 million - Rashard Lewis, sf, SEA
4. $7.80 million - Matt Harpring, sf, PHI
5. $7.62 million - Chauncey Billups, pg, MIN
6. $7.47 million - Keon Clark, c/pf, TOR
7. $7.03 million - Raef LaFrentz, c/pf, DAL
8. $7.00 million - Donyell Marshall, sf, UTA
9. $6.47 million - Mike Bibby, pg, SAC
10. $6.47 million - Rodney Rogers, pf, BOS
11. $6.26 million - Michael Olowokandi, c, LAC
12. $5.89 million - Troy Hudson, pg, ORL
17. $5.26 million - Ricky Davis, sg/sf, CLE
22. $3.79 million - Scott Padgett, f, UTA
27. $3.37 million - Bryon Russell, sf/sg, UTA
30. $3.29 million - Monty Williams, sf, ORL
34. $2.57 million - Zendon Hamilton, pf/c, DEN
37. $2.49 million - Greg Anthony, pg, MIL
47. $1.92 million - Kevin Willis, c/pf, HOU
48. $1.85 million - John Crotty, pg, UTA
55. $1.58 million - Trajan Langdon, sg, CLE
60. $1.39 million - Earl Boykins, pg, LAC
63. $1.30 million - Sam Mitchell, f, MIN
When he's not writing about the Sonics for Hoopsworld.com and SonicsCentral.com, Kevin Pelton usually has Excel open to check out NBA statistics. He can be reached via e-mail at email@example.com.